Acoustic interference suppression through speaker-aware processing

Patent metadata

Field	Value
Publication number	US-12477291-B2
Application number	US-202318386359-A
Country	US
Kind code	B2
Filing date	Nov 2, 2023
Priority date	Dec 9, 2022
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

Disclosed are systems, methods, and other implementations for acoustic interference suppression, including a method that includes obtaining a multi-source sound signal sample combining multiple sound components from a plurality of sound sources in a sound environment, with the plurality of sounds sources including one or more interfering sound sources produced by one or more loudspeakers in the sound environment, determining interfering sound characteristics for one or more sound signals that correspond to the one or more interfering sound sources, and suppressing at least one of the multiple sound components associated with the determined interfering sound characteristics for at least one of the one or more sound signals.

First claim

What is claimed is:

1 . A method for suppressing acoustic interference, the method comprising:
obtaining a sample of a multi-source sound signal, wherein the sample includes sound components from sound sources that are in a sonic environment, wherein the sound sources include an interfering sound-source that is produced by a loudspeaker in the sonic environment, and wherein the multi-source sound signal includes an interfering signal from the interfering sound source; determining sound characteristics of the interfering signal; causing a machine-learning system to receive sound data that represents the multi-source sound signal, the machine-learning system being an interference-suppressing machine-learning system that has been trained to suppress sound produced by the interfering sound source; and using the machine-learning system, suppressing the interfering signal that is in the sample of the multi-source sound signal.

2 . The method of claim 1 , wherein the interfering sound source includes audio from a voice assistant, the audio having been produced using a voice-assistant profile selected from among a plurality of voice-assistant profiles that are maintained by a text-to-speech voice-assistant system.

3 . The method of claim 1 , wherein determining sound characteristics for the interfering signal comprises selecting, from data storage, interfering sound characteristics that correspond to a selected voice-assistant profile.

4 . The method of claim 1 , wherein suppressing the interfering signal comprises causing the machine-learning system to process the sound data according to the sound characteristics and to produce output data that represents an output sound sample, wherein, in the output sample, the sound produced by the interfering sound source has been suppressed.

5 . The method of claim 1 , wherein suppressing the interfering signal comprises causing the machine-learning system to determine filter coefficients for a time-varying linear filter based on the sound data and the sound characteristics of the interfering signal and applying the filter coefficients thus determined to a representation of the sample of the multi-source sound signal to yield an interference-suppressed output sound-signal.

6 . The method of claim 1 , wherein suppressing the interfering signal comprises causing the machine learning system to generate predicted output in which the interfering signal has been suppressed.

7 . The method of claim 1 , wherein determining the sound characteristics of the interfering signal comprises receiving a sound signal produced by the interfering sound source and deriving the characteristics of the interfering signal from that sound signal.

8 . The method of claim 1 , wherein suppressing the interfering signal comprises causing the machine-learning system to receive characteristics of the interfering signal that have been derived from a sound signal provided by the interfering sound source and to use the characteristics for providing an output that represents the sound signal with the interference signal having been suppressed.

9 . The method of claim 1 , wherein determining the sound characteristics of the interfering signal comprises determining a vector representation for the interfering signal.

10 . The method of claim 1 , wherein determining the sound characteristics of the interfering signal comprises determining spectral characteristics for the for the interfering signal.

11 . The method of claim 1 , wherein determining the sound characteristics of the interfering signal comprises determining identifier data that identifies the interfering sound-source.

12 . The method of claim 1 , wherein suppressing the interfering signal comprises avoiding use of a reference signal for any of the sound sources to reduce echo effects caused by those sound sources.

13 . The method of claim 1 , wherein determining the sound characteristics comprises causing the machine-learning system to receive interference data, the interference data being data that represents the sound characteristics, causing the machine-learning system to receive sound data, the sound data being data that represents the sample of the multi-source sound signal, and causing the machine-learning system to produce output data that represents an output sound sample in which the interfering signal has been suppressed.

14 . The method of claim 1 , wherein determining the sound characteristics comprises causing an acoustic transducer to convert a raw sound sample into a time-domain sound signal, transforming the time-domain sound signal into a transformed domain representation, and extracting features from the transformed domain representation to produce the sound characteristics.

15 . The method of claim 1 , wherein determining the sound characteristics comprises causing an acoustic transducer to convert a raw sound sample into a time-domain sound signal, transforming the time-domain sound signal into a transformed domain representation, and deriving from the transformed domain representation one or more of: complex signal spectra features, spectral magnitude features, log spectral magnitude features, log mel spectra features, or mel-frequency cepstral coefficients.

16 . The method of claim 1 , wherein suppressing the interfering signal comprises causing the machine learning system to generate predicted output in which the interfering signal has been suppressed, wherein the predicted output comprises at least one of: a temporal representation of an output sound sample and a spectral representation of an output sound sample, wherein, in the output sound sample, the sample, the sound produced by the interfering sound source has been suppressed.

17 . The method of claim 1 , wherein the machine learning system includes an input layer and wherein the method further comprises providing the determined sound characteristics to the input layer.

18 . The method of claim 1 , wherein the machine learning system includes an intermediate hidden layer and wherein the method further comprises providing the determined sound characteristics to the intermediate hidden layer.

19 . An apparatus for suppressing acoustic interference, the apparatus comprising:
an audio acquisition section that obtains a sample of a multi-source sound signal that includes sound components from sound sources that are in a sonic environment, the multi-source sound signal including an interfering signal from an interfering sound source that is among the sound sources, wherein the interfering sound-source is produced by a loudspeaker in the sonic environment, and a controller that is in electrical communication with the audio acquisition section, the controller being an acoustic interference suppression controller that comprises a machine-learning system that has been trained to suppress sound produced by the interfering sound source, wherein the controller is configured to receive sound data that represents the multi-source sound signal and to suppress the interfering signal that is in the sample of the multi-source sound signal.

20 . A manufacture comprising a non-transitory computer readable medium that stores instructions for causing a machine to carry out steps of:
obtaining a sample of a multi-source sound signal, wherein the sample includes sound components from sound sources that are in sonic environment, wherein the sound sources include an interfering sound-source that is produced by a loudspeaker in the sonic environment, and wherein the multi-source soun

Assignees

• Cerence Operating Co

Inventors

• Faubel Friedrich
• Jungclaussen Jonas
• Haulick Tim
• Buck Markus

Classifications

• H04S2400/15

  Aspects of sound capture and related signal processing for recording or reproduction · CPC title

• H04S2400/11

  Positioning of individual sound objects, e.g. moving airplane, within a sound field (H04S2420/13 takes precedence) · CPC title

• H04S7/40

  Visual indication of stereophonic sound image · CPC title

• H04R2410/03

  Reduction of intrinsic noise in microphones · CPC title

• H04R1/08

  Mouthpieces; {Microphones;} Attachments therefor · CPC title